Estimating feels like a talent for numbers, so people who miss tend to decide they are bad at maths. The misses in roughly's own data point somewhere else. Players made the Hindenburg about a fifth of its real length, as the Hindenburg analysis shows, and moved Bronze Age objects about 1,500 years towards today, as the Bronze Age analysis shows. Neither is an arithmetic error. Both are a missing or a wrong anchor.
Every good estimate starts from an anchor you already own
Amos Tversky and Daniel Kahneman named the move in 1974: in numerical prediction, when a relevant value is available, people adjust from an anchor (Science, via PubMed).
The same paper says these shortcuts are usually effective and still lead to systematic and predictable errors. That is the whole catch: an anchor helps when it is relevant and hurts when it is arbitrary. A football pitch is a good anchor for the height of a rocket. A random starting size is a bad one. Every round of How big? starts at a wrong size, so the first job is to move away from it, towards something you know.
The reveal shows the anchor you should have used
If anchors do the work, the useful question after a miss is which anchor was missing. Every reveal on roughly ends with the way you could have found the answer, and the sample reveals on the game pages show the pattern.
| Game | Question | Anchor in the reveal |
|---|---|---|
| How big? | How tall is a Saturn V? | A 105 m football pitch, Big Ben at 96 m |
| How many? | How many of you tall is the Statue of Liberty? | Ten of you stacked reach 17 m |
| When? | When was the fax patented? | The telephone patent of 1876 |
| How likely? | Lifetime odds of lightning in the US | About 270 people a year, out of 330 million |
So the anchor is rarely the answer. It is a nearby number you can picture, and the arithmetic from there takes one or two steps. A few anchors do a lot of the work: in the When? data set, 60 of the 327 reveal paths work back from the Moon landing, the Great Pyramid or the Colosseum.
Anchors can mislead, so pick yours instead of taking one
The strongest objection comes from the same research: anchoring is a bias, so leaning on anchors should make estimates worse. It does, when the anchor is handed to you and has nothing to do with the question. The fix is to choose your own and check it against the order of magnitude.
The Hindenburg round shows both sides. Players were not stuck on the starting size: three in four moved away from it, downwards. The post's best explanation is that they swapped one anchor for another, the blimp over a stadium, which is about a quarter of the real airship. A better anchor, three Boeing 747s nose to tail, would have come close.
Anchors fail where you own none
In a field you have never seen at all, there may be nothing to start from. Then the honest move is the order of magnitude, 10, 100 or 1,000, and nothing finer.
Before the next guess, say the anchor first: one known thing, how many of it fit, then round hard. After the reveal, keep the anchor the sheet hands you, because it is the one you will need next time. For counting what is in front of you, Amountle trains the same move on piles in a glass.
Five sizes a day, and every reveal hands you the anchor.
Play today's How big?Sources 1
- Tversky and Kahneman 1974, Judgment under Uncertainty: Heuristics and Biases (PubMed) - adjustment from an anchor in numerical prediction, and the systematic errors these heuristics cause